Graph Convolution Point Cloud Super-Resolution Network Based on Mixed Attention Mechanism

Round 1

Reviewer 1 Report

 

This paper proposed “Graph Convolution Point Cloud Super-Resolution Network 2 Based on Mixed Attention Mechanism”. To improve the quality of the manuscript, I recommend following corrections.

1-     Introduction section needs more investigation of some latest and relevant work and add contribution at end of introduction and related section.

2-     Literature review not fully explored and what is so far work has been carried out in this area not fully covered. I suggest the authors to add related work section and come up with clear flaws in state of art approaches and justification of their work (I suggest a paper (Lidar Point Cloud Compression, Processing and Learning for Autonomous Driving).

3-     author fully exploit the global and local features of point cloud through the attention mechanism and improve the learning ability of the network, at the same time dense connected network (DCN) including multiple feature extraction modules (FEM) is employed to generate high quality feature maps. Each FEM is composed of a dilated graph convolution block (DGCB) and mixed attention mechanism What is your contribution, and how can we improve learning ability through these models?

4-     In order to make good use of global and local features of point cloud, the mixed attention mechanism is designed to integrate channel attention within DGCB to filter in ormation on the channel dimension and multi-head attention between DGCB to make the network focus on the more valuable information from the convolution layer of different depths. Clearly define depths range.

5-     Add more detail in the captions for all figures so that they are self-explanatory to the reader. 

6-        The experiments section needs more comparison with other research works.

7-        You should test your model with other datasets, add a separate experiments section, and compare it with further research and DGCM results.

8-     Even its not compulsory I suggest you upload your source code with revised manuscript.

9-     There are ambiguities and unclear meanings throughout the whole paper. Be more precise with your language.

 

 

 

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

Dear authors,

I have carefully read the manuscript entitled "Graph Convolution Point Cloud Super-resolution Network Based on Mixed Attention Mechanism."

The paper is well-written and interesting, and the results seem promising. Nevertheless, some issues should be solved before its acceptance for publication:

 

An English native speaker should revise the manuscript: major mistakes related to "has/have" are found; the article usage should be improved.

The main concern is related to References: there are few, and they are not very recent, which is not logical for a cutting-edge topic like this. Besides, the reference style does not fit the journal.

This reviewer would prefer another way of citing articles. For example, instead of "In [10] Li R. et al. propose...", I would write it: "Li et al. propose... [10]."

 

Kind regards.

 

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 3 Report

The parameters of the algorithm should be presented in a Table and give justification for each of the parameter.

What was the criterion for stopping the training

Repetition of “the test set in the PU1K dataset” 4.4 and the dataset section

Extend the discussion of the result presently it is more of result presentation and interpretation

What was the possible reason why the propose algorithm outperform the classical algorithms  

Discuss the implication of the results to theory and practice   

No need for restating the problem statement in the conclusion section……  

Data availability statement is missing

The paper is not up to date, the most recent reference in the paper is 2021, meaning that the authors have not consulted publications works on similar algorithms in 2022 and 2023. In summary, they are not aware of the happening in 2022 and 2023.

Unfortunately, more advance method has already been published

Liang, G., KinTak, U., Yin, H., Liu, J., & Luo, H. (2023). Multi-scale Hybrid Attention Graph Convolution Neural Network for Remote Sensing Images Super-resolution. Signal Processing, 108954.

Zhang, Q., Feng, L., Liang, H., & Yang, Y. (2022). Hybrid Domain Attention Network for Efficient Super-Resolution. Symmetry, 14(4), 697.

 

 

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 4 Report

Please find the file attachment.

Comments for author File: Comments.pdf

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 5 Report

These are JUST some of my comments and you need to go over the whole paper, rewrite and revise and highlight all changes.

“s it difficult to adapt to complex” awkward wording

“, we propose a novel graph convolutional point cloud super-resolu-tion network based on mixed attention mechanism (GCN-MA), which is composed of feature ex- 15 traction and point upsampling” what kind of sentence is that? It is very long and not readable.

You cannot just start a subsection by a figure, 3.1. Overview.

I am not able to read your paper. Highlight what your study has to offer in a separate figure so your readers could follow what you have done. Unfortunately, this paper could not be accepted unless the authors spend a significant time rewriting and revising it. Do NOT resubmit unless you spend a significant time.

Good luck

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 3 Report

The authors resolved all the issues.

Author Response

Thanks for your comments.

Reviewer 4 Report

Dear Authors,

Appreciate your efforts in addressing the concerns highlighted in the previous report and reflecting those modification in manuscript. Overall manuscript improved to convey the actual scope of proposed research work.

But in the beginning of abstract, authors mentioned that "Due to the limitation of terrain, weather or acquisition equipment is is challenging  to directly obtain dense and high-quality point cloud data by 3D scanning".  And proposed approach will introduced method to over come above challenges compared to state-of-art method. In this phrase, objective sounds more broader than actual scope of proposed work. Because presented evidence in results section will not support this claim. Presented results doesn't exclusively address issues associated with terrain limitation, weather or acquisition  equipment. If so, i would suggest to provide super sampled point cloud data of terrain, challenging weather, sparse point cloud from low resolution equipment. Else, rephrase the line clearly to depict actual scope of this research work. 

Author Response

Thanks for your comments. We have revised the first sentence of the abstract.

Reviewer 5 Report

well done!

Author Response

Thanks for your comments.